Health Monitoring in an Agent-Based Smart Home

fantasicgilamonsterΔιαχείριση Δεδομένων

20 Νοε 2013 (πριν από 3 χρόνια και 8 μήνες)

140 εμφανίσεις

Health Monitoring in an Agent
-
Based Smart Home

Diane J. Cook, Sajal Das, Karthik Gopalratnam, and Abhishek Roy

Department of Computer Science Engineering

University of Texas at Arlington

cook@cse.uta.edu



1. Introduction

We live in an increasingly connec
ted and automated society. We are investigating monitoring and
automation assistance in our most personal environment: the home. This integration of engineering and
life science builds upon UTA's MavHome project
[
6
]
, a home e
nvironment that perceives the state of the
home through sensors and intelligently acts upon the environment through controllers.

As Lanspery, et al. state, "For most of us, the word `home' evokes powerful emotions [and is] a
refuge"
[
12
]
. They note that older adults and people with disabilities want to remain in their homes even
when their conditions worsen and the home cannot sustain their safety.
Furthermore, t
he problems of
aging and disability are converging. Improvements
in medical care are resulting in increased survival into
old age, thus problems of mobility, vision, hearing, and cognitive impairments will increase. As the baby
boomers enter old age, this trend
will be magnified. By 2040, 23
% will fall into the 65+ ca
tegory
. An
AARP report [
1
]

strongly encourages increased funding for home modifications that can keep older adults
with disabilities independent in their own homes.

Our goal is to assist the elderly and individuals with disabil
ities by providing home capabilities that
will monitor health trends and assist in the inhabitant's day to day activities in their own homes. The
result will save money for the individuals, their families, and the state. We are seeking to meet this goal
using the MavHome smart home environment. MavHome is equipped with sensors that record inhabitant
interactions with many different devices, medicine
-
taking schedules, movement patterns, and vital signs.
We have developed algorithms that learn patterns of

activities from this data
, and

are
apply
ing

these
capabilities to health monitoring

in

the following
ways
:



Perform secure, context
-
aware collection of inhabitant health and activity data,



Use our data mining and prediction techniques to learn patterns in
collected data,



Identify trends that could indicate health concerns or a need for transition to assisted care,



Detect anomalies in regular patterns that may require intervention, and



Provide reminder and automation assistance for inhabitants.

By investigat
ing these issues we can offer the community an intelligent system with learning
algorithms that not only perform their individual tasks well, but also form a synergistic whole th
at is
stronger than the parts.


2.
The MavHome Smart Home

We define an intell
igent environment as one that is able to acquire and apply knowledge about its
inhabitants and their surroundings in order to adapt to the inhabitants and meet the goals of comfort and
efficiency. These capabilities rely upon effective prediction, decisio
n making, mobile computing, and
databases. With these capabilities, the home can control many aspects of the environment such as
climate, water, lighting, maintenance, and entertainment. Intelligent automation of these activities can
reduce the amount of

interaction required by inhabitants, reduce energy consumption and other potential
wastages, and provide a mechanism for ensuring the health and safety of the environment occupants.

MavHome operations can be characterized by the following scenario. To mi
nimize energy
consumption, MavHome keeps the house cool through the night. At 6:45am, MavHome turns up the heat
because it has learned that the home needs 15 minutes to warm to Bob's desired waking temperature. The
alarm sounds at 7:00am, after which the

bedroom light and kitchen coffee maker turn on. Bob steps into
the bathroom and turns on the light. MavHome records this manual interaction, displays the morning
news on the bathroom video screen, and turns on the shower. When Bob finishes grooming, th
e
bathroom light turns off while the kitchen light and display turn on, and Bob's prescribed medication is
dispensed to be taken with breakfast. Bob's current weight and other statistics are added to previously
collected data to determine health trends th
at may merit attention. When Bob leaves for work, MavHome
reminds Bob remotely that he usually secures the home and has not done so today. Bob tells MavHome
to finish this task and to water the lawn. Because there is a 60% chance of rain, the sprinklers

are run a
shorter time to lessen water usage. When Bob arrives home, the hot tub is waiting for him. Bob has had
a long day and falls asleep in the hot tub. After 40 minutes MavHome detects this lengthy soak as an
anomaly and contacts Bob, who wakes up

and moves on to bed.












MavHome's smart home capabilities are organized into a software architecture that seamlessly
connects needed components while allowing improvements to be made to any of the supporting
technologies. Figure

1

shows the arc
hitecture of a MavHome agent. Technologies are separated into four
cooperating layers. The Decision layer selects actions for the agent to execute. The Information layer
collects information and generates inferences useful for decision making. The Comm
unication layer
routes information and requests between agents. The Physical layer contains the environment hardware
including devices, transducers, and network equipment. The MavHome software components are
connected using a CORBA interface. Because co
ntrolling an entire house is a very large and complex
learning and reasoning problem, the problem is decomposed into reconfigurable subareas or tasks. Thus
Figure 1. MavHome agent architecture.

the Physical layer for one agent may in actuality represent another agent somewhere in the hierarch
y,
which is capable of executing the task selected by the requesting agent.

Perception is a bottom
-
up process. Sensors monitor the environment (e.g., lawn moisture level) and,
if necessary, transmit the information to another agent through the Communicati
on layer. The database
records the information in the Information layer, updates its learned concepts and predictions accordingly,
and alerts the Decision layer of the presence of new data. During action execution, information flows top
down. The Decisi
on layer selects an action (e.g., run the sprinklers) and relates the decision to the
Information layer. After updating the database, the Communication layer routes the action to the
appropriate effector to execute. If the effector is actually another ag
ent, the agent receives the command
through its effector and must decide upon the best method of executing the desired action. Specialized
interface agen
ts allow interaction with users

and external resources such as the Internet.

Agents can communicate
with each other using the hierarchical flow shown in Figure

1
. Several
smart home projects have been initiated elsewhere
, including Georgia Tech, MIT, University of Colorado
at Boulder, and industry labs
. MavHome is unique in combining technologies from
artificial intelligence,
machine learning,
and
databases to create a smart home that acts as an intelligent agent.


3.
Learning to Identify Significant Episodes

In order to maximize comfort, minimize cost, and adapt to inhabitants, a smart home must rely
upon
tools from artificial intelligence such as data mining and prediction. Prediction can be used to determine
the inhabitant's next action. Specifically, MavHome needs to identify repetitive tasks performed by
inhabitants that establish a baseline for
learning trends in behaviors, detecting anomalies, and determining
repetitive tasks worthy of automation by the home. The home can make this prediction based solely on
previously
-
seen inhabitant activities and the current state of the inhabitant and the h
ouse.

A smart home inhabitant performs various routine activities, which may be considered as a sequence
of events, with some inherent pattern of recurrence. This repeatability leads us to the conclusion that the
sequence can be modeled as a stationary st
ochastic process. We can then perform inhabitant action
prediction by first mining the data (using ED) to identify sequences of actions that are regular and
repeatable enough to generate predictions, and by second using a sequence matching approach (Acti
ve
LeZi) to predict the next action in one of these sequences.

Our Episode Discovery (ED) data mining algorithm is based on the work of
Agrawal and Srikant

[
2
]

for mining sequential patterns from time
-
ordered transactions. We m
ove an examination window through
the history of events or inhabitant actions, looking for episodes (sequences) within the window that merit
attention, or
significant episodes
. Each candidate episode is evaluated using the Minimum Description
Length (MDL)

principle. The MDL principle favors patterns that can be used to minimize the description
length of a database by replacing each instance of the pattern with a pointer to the pattern definition. A
detected regularity factor (daily, weekly, or other
autom
atically
-
detected
time frame) further compresses
the data because the sequence can be removed without storing a pointer to the sequence definition, and
thus increases the value of a pattern. Deviations from the pattern definition in terms of missing event
s,
extra events, or changes in the regularity of the occurrence add to the description length because extra bits
must be used to encode the change, thus lowering the value of the pattern. The larger the potential
amount of description length compression a

pattern provides, the greater the impact that results from
automating the pattern.

Our ED algorithm successfully identified daily and weekly patterns in synthetic data based on the
MavHome scenario described earlier. We also used ED to mine data that was

collected in the MavHome
lab
environment from six students during the spring of 2003. The dataset contains 618 interactions that
are members of patterns occurring once a week, multiple times a week, and randomly. ED successfully
identified the patterns
of three of the inhabitants as weekly significant episodes, and marked which of the
618 interactions contributed to the significant episodes
[
10
]
.

The knowledge that ED obtains by mining the user action history can be used in a v
ariety of ways.
First, the mined patterns provide information regarding the nature of activities in the home, which can be
used to better understand lifestyle patterns and aid in designing homes and devices for the home. Second,
the significance of a cur
rent event as a member of a discovered pattern can be used in controlling the
home, to determine whether this task is worth attempting to automate. Third, knowledge of the mined
sequences can improve the accuracy of predicting the next action, by only per
forming prediction for
events known to be part of a common pattern. We demonstrate the ability of ED to perform the third task,
improving the accuracy of prediction algorithms, by adding the mined results as a preprocessor to two
prediction algorithms. A
ction sequences are first filtered by the mined sequences. If a sequence is
considered significant by
ED
, then predictions can be made for events within the sequence window.

To test the filtering capabilities of ED, we coupled it with the IPAM sequentia
l predictor
[
7
]

and a
back
-
propagation neural network (BPNN). We created a sequence of 13,000 actions based on five
randomly
-
generated scenarios, a situation in which these algorithms by themselves may not perform well.
ED dis
covered 14 episodes in the data sets, and appreciably improved the accuracy of both algorithms
across all five scenarios, as can be seen in Table

1
.
U
sing ED, we improve the accuracy of the prediction
algorithms by reducing the total number of incorrect p
redictions

that can lead

to inaccuracies in learned
health trends, detected anomalies, and automated patterns.

Scenario

1

2

3

4

5


Average

Events

12
958

12
884

12
848

13
058

126
68

128
83

Episode

Candidates

57
45

56
08

56
19

56
55

549
6

562
5

Significant

Episodes

13

13

13

13

13

13

IPAM

Percentage

Correct

39
%

42
%

43
%

40
%

41
%

41
%

IPAM+ED

Percentage

Correct

77
%

84
%

69
%

73
%

65
%

74
%

BPNN

Percentage

Correct

62
%

64
%

66
%

62
%

64
%

64
%

BPNN+ED

Percentage

Correct

84
%

88
%

84
%

84
%

88
%

86
%

Processing

Time
(s)

11

9

10

9

9




10


Table 1. Prediction improvement results.


4.
Learning to Predict Inhabitant Actions

Prediction is an important component in a variety of domains in artificial intelligence and machine
learning, which

allows intelligent systems to make more info
rmed and reliable decisions. Certain
domains require that prediction be performed on sequences of events that can typically be modeled as
stochastic processes. Especially common is the problem of sequential prediction: given a sequence of
events, how do

we predict the next event based on a limited known history. This is true, for example,
when predicting inhabitant actions in a smart environment such as MavHome. Prediction can be
performed of upcoming inhabitant activities based on observed past activi
ties.

Our prediction algorithm is based on the LZ
78 text compression algorithm [
14
]. Good text
compression algorithms have also been established as good predictors. According to information theory,
a predictor with an order (s
ize of history used) that grows at a rate approximating the entropy rate of the
source is an optimal predictor [
8
].

LZ78 processes an input string of characters, which in our case is a string representing the history of
inhabita
nt actions interacting with devices in the home. The prediction algorithm parses the input string
x
1
,x
2
,…,x
i

into
c(i)

substrings, or phrases,
w
1
, w
2
, …, w
c(i)

such that for all
j>0
, the prefix of the substring
w
j

(i.e., all but the last character of
w
j
)
is equal to some
w
i

for
1<i<j
. Because of the prefix property used
by the algorithm, parsed substrings can be maintained in a trie along with frequency information.

Consider the sequence of input symbols
aaababbbbbaabccddcbaaa
.
An LZ78 parsing of this i
nput
string would yield the following set of phrases:
a,aa,b,ab,bb,bba,abc,c,d,dc,ba,aaa
.
As described above,
this algorithm maintains
statistics for all contexts seen within each phrase w
i
.

For example, the context
a

occurs 5 times (at the beginning of th
e phrases
a, aa, ab, abc, aaa
), the context
bb

is seen 2 times
(
bb,bba
), etc. These context statistics are stored in a trie.

Because it is designed as a text compression algorithm, LZ78 requires some enhancements to
perform effective prediction. For examp
le, we can see that the amount of information being lost across
phrase boundaries grows with the number of possible states in the input sequence. In our Active LeZi
(ALZ) algorithm, we enhance LZ78 to recapture information lost across phrase boundaries.
Specifically,
we maintain a window of previous
ly
-
seen symbols, with a
size

equal to the length of the longest phrase
seen in a classical LZ78 parsing. The reason for selecting this window size is that the LZ78 algorithm is
essentia
lly constructing an appr
oximation to an

order
-
k

Markov model, where
k
is equal to the length of
the longest LZ78 phrase seen so far.
ALZ

builds a better approximation to the order
-
k

Markov model,
because it has captured information normally lost across phrase boundaries. As a r
esult, we gain a better
convergence rate to optimal predictability as well as achieve greater predictive accuracy. Figure
2

shows
the trie formed by the Active LeZi parsing of the input sequence
aaababbbbbaabccddcbaaa
.

To perform prediction, the algorithm

calculates the probability of each symbol (action) occurring in
the parsed sequence, and predicts the action with the highest probability. To achieve optimal
predictability, we must use a mixture of all possible order models (phrase sizes) when determini
ng the
probability estimate. Active LeZi performs a second refinement of the LZ78 algorithm to combine this
predictive information. To accomplish this, we incorporate ideas from the Prediction by Partial Match
(PPM) family of predictors, which has been a
pplied to great effect in the mobility prediction work of
Bhattacharya and Das [
6
].








PPM algorithms consider different order Markov models to build a probability distribution. This
blending strategy assigns greater weigh
t to higher
-
order models, in keeping with the advisability of
Figure
2
. Tr
ie formed by sample history sequence.

a(10)

b(8)

c(3)

d(2)

b(1)

c(1)

d(1)

c(1)

d(1)

c(1)

b(1)

c(1)

b(4)

a(3)

b(3)

a(5)

d(1)

b(1)

c(1)

a(2)

a(1)

c(1)

a(1)

d(1)

Λ

a(2)


making the most informed decision. We employ the PPM strategy of
exclusion

[
3
]

to gather information
from all of the
1..k

order models in assigning the next symbol i
ts probability value.

As an example, consider our example string
aaababbbbbaabccddcbaaa
, ending in the phrase
aaa
.
Within this phrase, the contexts that can be used for prediction are all suffixes within the phrase, except
itself (i.e.,
aa
,
a
, and the nul
l context). From Figure
2

we see that an
a

occurs two out of the five times
that the context
aa

appears, the other cases producing two null outcomes and one
b
. Therefore the
probability of encountering
a

at the context
aa

is 2/5, and we now fall back (esc
ape) to the order
-
1 context
(i.e. the next lower order model) with probability 2/5. At the order
-
1 context, we see an
a

five out of the
ten times that we see the
a

context, and of the remaining cases, we see two null outcomes. Therefore we
predict symbol
a

at the order
-
1 context with probability 5/10, and escape to the order
-
0 model with
probability 2/10. At the order 0 model, we see the
a

ten out of 23 symbols seen so far, and we therefore
predict
a

with probability 10/23 at the null context. The
blended

p
robability of seeing
a

as the next
symbol is therefore 2/5 + 2/5{5/10 + 2/10(10/23)}.

U
sing the synthetic data generator, we created thirty days of activities using six different scenarios
and test the ability of Active LeZi to generate correct predictions
, given a model built from all previous
events, for the next 100 events. In the first experiment, the data consists only of events drawn from the
scenario definitions. The predictive accuracy in this case converges to 100%, as shown in Figure
3
. For
the

second experiment we introduce noise in the form of events not part of any scenario and variations in
event orderings. In this case, the predictive accuracy does improve with the amount of training data, but
converges to only 86% accuracy.







ALZ Performance - I
Repetitive Data without Noise
0
20
40
60
80
100
0
500
1000
1500
2000
2500
Number of Training Instances
% Prediction Accuracy

0

10

20

30

40

50

60

70

80

90

100

0

500

1000

1500

2000

Number of Training Instances

% Prediction Accuracy

ALZ Performance


II

Data with variation

Figure
3
. Prediction performance on sample smart home data.



We al
so tested the ability of Active LeZi to perform prediction on the real data collected in the
MavHome environment.
T
he accuracy of the model does improve with the amount of training data, but
only converges to 48%. However, this represents an improvement
over random choice, which for this
data would result in an average accuracy of 2%. Combining ALZ with ED yields a 14% improvement in
predictive accuracy for this data.



5.
Using a Smart Home to Assist Elderly and
People with Disabilities

The data mi
ning, prediction, and multiagent technologies available in MavHome can be employed to
provide health care assistance in living environments. Specifically, models can be constructed of
inhabitant activities and used to learn activity trends, detect anomali
es, intelligently predict possible
problems and make health care decisions, communicate with caregivers, and provide automation
assistance for inhabitants with special needs.

A variety of approaches have been investigated in recent years to automate caregi
ver services. Many
of the efforts offer supporting technologies in specialized areas, such as using computer vision techniques
to track inhabitants through the environment and specialized sensors to detect falls or other crises. Some
special
-
purpose pred
iction algorithms have been implemented using factors such as measurement of
stand
-
sit and sit
-
stand transitions and medical history, but are limited in terms of what they predict and
how they use the results. Remote monitoring systems have been designed
with the common motivation
that learning and predicting inhabitant activities is
a
key
to effective automated

health monitoring, but
very little work has combined the remote monitoring capabilities with prediction for the purpose of health
monitoring. Som
e work has also progressed toward using typical behavior patterns to provide reminders,
particularly useful for the elderly and patients suffering from various types of dementia
[
9
,
11
,
13
]
.

Our project differs from these earlier explorations in that we are combining capabilities in the areas
of data collection, remote monitoring, prediction, data mining, and knowledge engineering to provide
predictive health monitoring assist
ance for inhabitants with disabilities and caregivers. Instead of
designing one specialized component, we propose to show that a smart environment can accomplish all of
the tasks needed to identify patterns indicating or predicting a sudden or slow change

in health status, to
supply caregivers with periodic or emergency information, and to provide inhabitants with needed
automation assistance.

Capability 1: Perform Secure, Context
-
Aware Data Collection
.
The first capability we are
developing is that of r
emote collection of long
-
term activity and health status data. Data will consist of
monitored inhabitant activities, vital signs, and interactions with the environment, and will be collected
using context
-
aware technology.

Our smart home environment is cu
rrently equipped to gather the following types of information: 1)
inhabitant usage of any electrical device in the home, 2) usage of water in the home and amount, 3)
temperature settings, 4) inhabitant weight, 5) inhabitant movement throughout the home, 6
) prescribed
and actual medicine dispensing schedule, 7) time, duration, and intensity of exercise, and 8) use of food
items in kitchen. Active sensors including wearable vital sign monitors can be integrated to further refine
the model.

We
are designing

efficient algorithms for collecting real
-
time data that are context
-

and/or situation
-
aware. For example, the inhabitant's current activity (e.g., cooking vs. watching television) or location in
the environment (e.g., bedroom vs. navigating the stairs) ca
n affect the choice of sensors to use, and thus
represent a defined context. The proposed data mining and prediction algorithms are highly scalable, a
desirable feature when numerous tiny, portable sensor devices are involved. Furthermore, the algorithms

create personal profiles and hence provide customized solutions to individual patients. There may also be
a need for ad hoc collaboration of various entities in the system as the emergency need arises. These
problems can be elegantly tackled with the hel
p of pervasive computing and communications technology.

Dynamic discovery of services and information (e.g. data fusion from heterogeneous sources) is an
integral part of any complex system that makes proactive decisions and supports context / situation
-
aw
are
computing. A service may be provided by an active or passive device like a sensor, a piece of software, a
database, a communication channel or network, a person, or a combination thereof. Many of the sensors
and controllers will be connected to each
other or a main computer through a wireless connection.
Therefore, service providers must also be able to authenticate, authorize and account for such services,
satisfying the desired quality of service requirements such as communication link bandwidth, e
nd to end
delay, jitter, etc. for transmitting vital multimedia data and images. We are undertaking a novel decision
-
theoretic approach to manage scare resources in wireless mobile and sensor networks based upon our
prior work in this area.

Use of sensors

in smart homes and on elderly or people with disabilities is critical for collecting,
storing, and processing appropriate data intelligently and in a timely manner. Also important is secure
data transmission, routing, sharing, and authentication over the

wireless sensor network and wireline
network to care providers.

Data collection and processing needs to be backed by an infrastructure that allows anytime,
anywhere reliable access to data sources (e.g., national data bases or vital records). This is cruc
ial for the
purpose of data mining, intelligent decision making and profiling. We will address the challenges of
developing means to (i) locate relevant information securely, efficiently and transparently, (ii) extract,
process, and integrate relevant info
rmation efficiently and securely, and (iii) interpret and communicate
the processed information intelligently and seamlessly.

Moreover, wireless sensor and pervasive networks are highly vulnerable to various security attacks
due to the nature of the medium

(susceptible to eavesdropping and traffic jamming). An important
challenge we will address is to develop authentication and key management protocols among various
sensor nodes to set up a secure information sharing and communication. This entails designi
ng secure,
robust and energy
-
efficient routing protocols in sensor networks. A secure routing protocol should also
protect the integrity and authenticity of routing messages, and prevent attackers from modifying them or
injecting harmful messages in the ne
twork.

Privacy of the information is highly required since personal information will be collected during the
monitoring phase. This important issue will be addressed with the help of (i) proper compression and
encoding algorithms so that only the device eq
uipped with them will be able to see the content, (ii) a
multilevel access scheme for stored information, and (iii) protected copying and transferring of the stored
information with matching keys only. These strategies will be implemented to protect the p
rivacy of vital
and personal data with a high level of confidence.

Capability 2: Identify Trends in Long
-
Term Data
.
The second health monitoring capability uses our
data mining and prediction techniques to identify trends in long
-
term data. Trends in ti
me
-
varying data
can be discovered and predicted using the same data mining and prediction techniques described earlier.
Instead of capturing individual events and readings to store in the trie and use for prediction, changes in
these values over varying p
eriods of time will be captured, stored, and predicted. In particular, for each
feature of interest, the qualitative change in value (increase, decrease, same) and quantitative relative
change (numeric difference) will be recorded. Because the chosen tim
e steps are limitless in value and
yet critical to learning trends, we will allow the user to determine the time step of choice (daily, weekly,
monthly, etc.) and use auto
-
correlation techniques to automatically determine the time step most
indicative of a

temporal trend in the data.

The user can select features to monitor for trend analysis, including 1) change in mobility (schedule,
time, room location, total movement), 2) change in amount of exercise, 3) change in deviation from
prescribed medicine sched
ule, 4) change in amount of smart home requested assistance (reminders or
automation assistance), 5) change in nutrition (types of food, number of calories), and 6) change in
amount and types of activities.

Using the ED algorithm, long
-
term trends can be d
iscovered from the raw data. The algorithm can
also be used to search for changes of a type of duration indicated by the user. In addition, the prediction
algorithm can be used to predict the upcoming changes in these features for the next time step. Fi
nally,
we can use data from individuals assessed by practitioners to learn classes of patient types, such as
patients who require a move to an assisted care facility. Time
-
varying data captured for a particular
individual can then be classified based on t
he learned models.

The result of this technology can be used in a number of ways. For patients recovering from an
illness or accident, the algorithms can be used to determine whether they are regaining strength and
vitality at an acceptable rate. For pat
ients at risk of various types of accidents and health risk situations,
the trends can be used to determine the current health risk and predicted short term health risk. These
analyses can aid a caregiver in deciding whether the individual requires a chan
ge in medication, activity
schedule, care scheduling, or environment.

Capability 3: Detect Anomalies in Current Data
.
Another direct outcome of our work in data
mining prediction is the ability to detect anomalies in collected data. For the purpose of o
ur work, we use
the intuitive notion of an anomaly as
a s
urprising or unusual occurrence
. With this notion in mind, the
anomaly value of each monitored activity or captured feature value can be collected and used to provide
health status information.

An i
mportant consideration in any anomaly
-
detection system is the regularity, or predictability, of
the data. Generally, the more predictable the data, the easier it is to detect anomalies. As a result, the
anomaly value of each data point is influenced firs
t by the degree of membership of the current state in a
significant episode, second by the significance value of the episode itself, and third by the deviation of the
observed activity from the predicted activity given the known episode.

The degree of memb
ership of the current state in any of the discovered significant episodes is
calculated as the probability of occurrence of the episode given the immediate history of observed
activities. A probability is returned for each episode. This probability is th
en multiplied with the
significance of the episode itself to calculate the regularity of the data context. The episode yielding the
highest regularity value, episode i, represents the baseline for the anomaly calculation. Once membership
in episode i is
established, the ALZ trie is used to provide the probability of the next predicted event e. If
the event that actually occurs has a very low probability given the context, the resulting event constitutes
an anomaly. If the regularity of the context is hi
gh and the probability of the observed event is low, this
event should be flagged as an anomaly. Thus the anomaly value of observed

event e can be calculated as
)).
(
,
|
(
)
(
*
)
,
(
e
IHistory
Episode
e
P
Episode
ce
Significan
Episode
e
Member
A
i
i
i


Anomalies can be be handled in a manner appropriate to the nature of
the event and anomaly value.
The criticality of events can be set by the user. For example, an anomaly in predicted interactions with
the house will typically have a low criticality and thus simply be recorded as an anomaly or prompt a
reminder of the in
habitant of their normal routine. For a more critical anomaly, such as a sudden
disruption in vital signs, the home will first attempt to contact the inhabitant. If contact cannot be made,
the home will contact the caregiver. Web cameras placed througho
ut the house can be used by the
caregiver to check on the inhabitant in such a situation and intervene with a medical difficulty.

Capability 4: Design Reminder Assistance System
.

An additional benefit that our smart home data
mining and prediction capabi
lities can offer is to provide a reminder system for inhabitants. Reminders
can be triggered by two situations. First, if the inhabitant queries the home for his/her next routine
activity, the activity with the highest probability will be given based on
the ALZ prediction. Second, for
episodes with a high degree of significance and criticality, if the inhabitant deviates from the normal
routine (creates an anomaly), the house will initiate contact with the inhabitant to remind him/her of the
next expecte
d activity. Such a reminder service will be particularly beneficial for individuals suffering
from dementia. The reminder can enable the individual to feel secure about performing daily activities,
and prevent accidents such as forgetting to turn off the

water in the bathtub or keeping the doors unlocked
when leaving the house.

Capability 5: Provide Automation Assistance
.
Intelligent automation of home activities can reduce
the amount of manual home interaction required by inhabitants. In earlier work,

we have developed a
decision
-
making algorithm to automate activities in the house
[
5
]
. Selection of actions for the house to
take is based on suggestions from a reinforcement learning algorithm, which minimizes necessary manua
l
intervention with the house and reduces actions with costly effects. The decision maker relies on input
from ALZ to suggest the typical next inhabitant action, and uses results from the ED data mining
algorithm to segment the learning space and improve
the quality of the learned results.

Users can specify specific activities they would like the house to automate (temperature control,
control of hard
-
to
-
reach devices, etc.). MavHome will then automate the activities based on learned
preferences and actio
n patterns. In addition, if the inhabitant does not respond to a reminder of a critical
event, MavHome will automate an activity to ensure safety of the individual. Such actions include
shutting off the bathwater, turning off the stove, or locking up the

house when it is empty.


6.
MavHome Status

Initial algorithms are in place for data mining, mobility and action prediction, decision making, and
automation. These capabilities are being used in the MavHome lab (MavLab) and apartment (MavPad).
Figure

3

shows lights in the entryway (top left) and on Ryan's desk (bottom left) turning on in reaction to
inhabitant activities (these automated actions are also reflected in the
ResiSim 3D
simulator). We will
next collect health
-
specific data and test it using
volunteers living in MavPad and MavHome, as well as
recruited residents of the C.C. Young Retirement Community in Dallas, Texas.











Figure 4. MavHome lab environment.





References

1.

AARP, These four walls… Americans 45+ talk about home and commun
ity, 2003.

2.

R. Agrawal and R. Srikant, Mining sequential patterns. Proceedings of the Conference on Data
Engineering, pages 3
-
13, 1995.

3.

T. C. Bell, J. G. Cleary, and I. H. Witten,
Text Compression
, Prentice Hall Advanced Reference
Series
, 1990
.

4.

A. Bhattach
arya and S. K. Das, LeZi
-
Update: An information
-
theoretic framework for personal
mobiity tracking in PCS networks. ACM/Kluwer Wireless Networks Journal, 8(2
-
3):121
-
135,
2002.

5.

D. J. Cook and M. Youngblood, Smart homes. To appear in
Encyclopedia of Human
-
C
omputer
Interaction
, 2003.

6.

S. Das, D. J. Cook, A. Bhattacharya, E. O. Heierman, III, and T.
-
Y. Lin, The role of prediction
algorithms in the MavHome smart home architecture. IEEE Wireless Communications, 9(6):77
-
84, 2003.

7.

B.D. Davison and H. Hisrch, Proba
bilistic online action prediction. Proceedings of the AAAI
Spring Symposium on Intelligent Environments, 1998.

8.

M. Feder, N. Merhav, and M. Gutman, Universal prediction of individual

sequences. IEEE
Transactions on Information Theory, 38(4), 1992.

9.

T. T. F
ulmer, P. H. Feldman, T. S. Kim, B. Carty, M. Beers, and M. Putnam, An intervention
study to enhance medical compliance in community
-
dwelling elderly individuals.
Journal of
Gerontological Nursing
, 25(8):6
-
14, 1999.

10.

E. Heierman and D. J. Cook, Improving h
ome automation by discovering regularly occurring
device usage patterns. Proceedings of the International Conference on Data Mining, 2003.

11.

H. Kautz, L. Arnstein, G. Borriello, O. Etzioni, and D. Fox, An overview of the assisted cognition
project.
Proceed
ings of the AAAI Workshop on Automation as Caregiver
, 2002.

12.

S. Lanspery, J. J. Callahan, Jr., J. R. Miller, and J. Hyde, Introduction: Staying put
. In

Staying
Put: Adapting the Places Instead of the People

(S. Lanspery and J. Hyde, eds.),
pages 1
-
22,
Ba
ywood Publishing Company, 1997.

13.

M. E. Pollack, C. E. McCarthy, I. Tsamardinos, S. Ramakrishnan, L. Brown, S. Carrion, D.
Colbry, C. Orosz, and B. Peintner, Autominder: A planning, monitoring, and reminding assistive
agent.
Proceedings of the Internationa
l Conference on Intelligent Autonomous Systems
, 2002.

14.

J. Ziv and A. Lempel, Compression of individual sequences via variable rate coding. IEEE
Transactions on Information Theory, IT
-
24:530
-
536, 1978.